Bandwidth to the World Project Status 
Les Cottrell, 10/27/01

We now have iperf running on about 21 hosts. The aggregate throughput
from SLAC to these hosts is about 1.3Gbps. There is a lot of concern
about the reliability of the iperf servers. Typically they die on a host
once a day. A lot of effort has gone into detecting and automatically
restarting them. Multiple strategies are necessary due to the non uniform
configuration of te remote hosts. The preferred is to use ssh to automatically
start and stop the servers, if this is not possible then the servers are
left running, and email sent taht they need a manual restart. We have also
had to take account of hosts updating OSs causing the installed version
of iperf to fail.

The OC12 from from ESnet to
SLAC was installed. However, ESnet still needs to configure things before 
turning it over to SLAC, so we are still running at OC3. The effect of the
OC12 on the iperf throughputs will be interesting, especially for the ESnet
sites. For the current throughputs, see 
http://www.slac.stanford.edu/comp/net/iperftests-html/International-Iperf-Tests.html 

Les is working with our collaborators at FNAL (need machine & account), 
LANL (need an account), GSFC (need to try out account and configure host),
and Rice (need to understand how to make ssh work)
to either get the remaining sites working or get the final 
configuration (e.g. host, ports allowed etc.) at the sites set up.
During the last week, NERSC & UFL have each added a 2nd host, and
Wisconsin upgraded the host so we are now testing the new throughputs.
We now get over 77Mbps instead of 9Mbps. 
U Michigan (Thomas Hacker) have expressed interest in joining the
collaboration and will provide a host.

Connie & Les have now automated the running of the bbcp tests to many of the sites.
Bbcp has been extended to allow a timeout option.
In order to accomodate more idiosyncracies of various sites we have also
extended the configuration file. In addition we have written test programs
to quickly test the remote hosts configurations etc.

The two new Dells's have been configured by Steffan Luitz. 
With jumbo frames and 2 NICs (on different subnets) we can achieve almost
2 Gbit/s over TCP (990 + 988). With 1500 byte MTUs we max out around 1.4
Gbit/s. The two iperfs were running at the same time, not sequentially.
The third Dell from last year (Antonia) has now had the 100 Mbps bottleneck
(used for QBSS testing) removed.

Jerrod Williams now has the ping displays working with the new version of the
Universal Time History Java servlet (from Tony Johnsons)
that supports transparent gifs,
improved font and frame control and placement, plus the twinkle
displays, see
http://www-iepm.slac.stanford.edu/monitoring/bulk/sc2001/iperf-map.gif

Connie is working on a web page to display throughputs from iperf, bbcp and 
derived form ping RTT & loss.

We will hand over the hosts to shipping this coming Monday.

The web page for the project is at:
http://www-iepm.slac.stanford.edu/monitoring/bulk/sc2001/